Fault Analysis and Evaluation of a True Random Number Generator Embedded in a Processor

True Random Number Generators have many uses, in particular they play a key role in security applications and cryptographic algorithms. Our interest lies in the quality of their generated random numbers. More specifically, for such utilizations, a slight deviation of the numbers from a “perfect” behavior can have disastrous consequences. It is then necessary to devise schemes for the testing of these generators in order to detect non-random properties of their numbers. Moreover, one should consider them from an attacker point of view and use any means to try to perturbate their good functionnality. In this article we describe such experiments and several standard statistical tools for the generators testing. We also present experimental results obtained through the study of a generator embedded in a processor in order to illustrate our methodology. We show that its pertubation leads to the apparition of dangerous deviations in its numbers distribution.